Comment by paxys
5 months ago
OpenAI - "Accuracy is a huge problem with LLMs, so we gave ChatGPT an internal thought process so it can reason better and catch mistakes."
You - "Amazing, so we can check this log and catch mistakes in its responses."
OpenAI - "Lol no, and we'll ban you if you try."
Yes this seems like a major downside especially considering this will be used for larger complex outputs and the user will essentially need to verify correctness via a black box approach. This will lead to distrust in even bothering with complex GPT problem solving.